CoreWeave’s durability test: customer research reveals AI cloud gaps

CoreWeave has spent the last two years as the darling of AI infrastructure, riding Nvidia’s GPU wave to a high valuation and a reputation for outmaneuvering hyperscalers on speed and flexibility. Now, ahead of its Fully Connected conference, the company is confronting a harder question: Can it turn temporary GPU rentals into a durable cloud business? New research from Qualitate, commissioned by SiliconANGLE, suggests the answer isn’t obvious.

The study, based on 13 in-depth interviews with enterprise buyers and operators, reveals a gap between CoreWeave’s strengths and what customers actually value long-term. Buyers praise the company’s ability to deliver scarce hardware quickly—critical during periods of high demand—but express skepticism about its staying power. That tension defines CoreWeave’s next phase.

The company’s early success stemmed from a simple bet: that AI startups and enterprises would pay a premium for immediate access to hardware, even if it meant bypassing AWS, Google Cloud, or other major providers. It worked. CoreWeave’s growth has been fueled by workloads that couldn’t wait for traditional cloud allocations. When we covered its recent test of new Nvidia hardware, the focus was still on raw performance and time-to-deployment. But the Qualitate research suggests those metrics may not be enough.

Enterprise buyers are now weighing risks. First, cost: pricing, while competitive for short-term bursts, becomes harder to justify for sustained use. Second, lock-in: customers worry about migrating off the platform if the company pivots or changes direction. Third, ecosystem: unlike major cloud providers, CoreWeave lacks the tooling and integrations that enterprises rely on for compliance, monitoring, and scaling. One interviewee put it bluntly: “I can’t run my production AI on a cloud that doesn’t have proper enterprise controls.”

CoreWeave appears to be responding by expanding its capabilities. Reports suggest it is adding more software features and leaning into partnerships, which could help address some ecosystem concerns. But these are incremental steps. The bigger challenge is proving it can retain customers beyond the era of hardware scarcity.

That era may be shifting. Supply has stabilized, and hyperscalers are aggressively courting AI workloads with their own infrastructure. Recent moves by competitors, like Google’s Project Suncatcher, highlight that the market isn’t standing still. If CoreWeave wants to avoid becoming a footnote in the AI infrastructure story, it needs to show it can do more than provide hardware—it needs to build a cloud.

The Fully Connected conference will likely focus on product announcements and partnerships. But the real test isn’t what CoreWeave says next week; it’s whether its customers keep coming back. The Qualitate research suggests they’re waiting for a reason to.

Sources: siliconangle.com